Documentation
Company Brain Guide
Everything you need to build, maintain, and share your company's governed knowledge base.
What is Company Brain?
Company Brain is a governed knowledge engine that aggregates information from your business tools — CRMs, documents, Slack, strategy platforms — and distils it into a single, verified source of fact. That knowledge is then made available to every AI tool your team uses, automatically, via the Model Context Protocol (MCP).
Instead of every employee prompting their AI assistant with background context from scratch, Company Brain loads your company's verified knowledge into every AI session automatically. One setup, continuous benefit.
Connect
Link your tools — CRM, Docs, Slack, strategy platforms — as knowledge sources.
Distil
AI extracts candidates. You review them. Approved ones become verified Facts.
Deliver
Every AI session — Claude, Cursor, ChatGPT — starts with your company context loaded.
How to Get Started
Get Company Brain serving context to your AI tools in four steps.
Upload your first knowledge
Go to Data Upload and paste a company overview, upload a document (PDF, DOCX, TXT), or connect an integration. Claude AI extracts facts automatically.
Review the proposal queue
Extracted facts land in your Review Queue as Proposals. Review each one — press V to verify or D to dismiss. Approved proposals become Facts.
Connect your AI tool
Go to MCP Connect, generate an API key, and follow the 3-step setup for your AI client (Claude Desktop, Cursor, Devin Desktop, ChatGPT, and more).
Share with your team
Go to Team, generate a company key, and share the setup link. Teammates click it, pick their AI tool, and connect — no Company Brain account needed.
Tip — Enable Brain Agents
Once you have initial Facts, enable Brain Agents to keep your knowledge current automatically. Agents monitor your integrations and propose updates as things change.
Core Concepts
Company Brain is built around five interconnected data types. Understanding how they relate makes the whole system click.
The canonical facts about your company. Verified by humans, served to AI. Every fact has a category, confidence score, source, and effective date range. Facts are immutable once approved — edits create new versions.
Fact candidates awaiting review. Generated by the AI extraction pipeline, autonomous agents, or manual submission. Each proposal shows its confidence score, supporting evidence, and any conflicts with existing Facts.
Source excerpts and data points that back up proposals and facts. Every approved Fact carries traceable evidence showing exactly where the fact came from.
Important observations — market shifts, competitive moves, customer feedback, internal events. Signals don't become Facts directly but inform them. Agents monitor signals and generate proposals when patterns emerge.
Knowledge Workflow
Information flows from raw sources through extraction, human review, and into verified Facts served to AI.
Source — A document, integration, or pasted text. Registered in your sources list with a trust level.
Extraction — Claude AI runs two passes: structured facts (names, numbers, dates) and strategic context (mission, positioning, process). Runs on upload or via agents.
Proposal — Each extracted fact becomes a proposal with a confidence score (0–1). High confidence + no conflicts = auto-approved if you have that setting enabled.
Human Review — Your team reviews proposals in the Review Queue. Approve to create a Fact, dismiss to reject.
Fact — Verified, versioned, categorised. Returned to every connected AI client via MCP tools.
AI Context — When a teammate opens Claude/Cursor/ChatGPT with Company Brain MCP connected, the relevant Facts load automatically.
Conflict Detection
When a new proposal contradicts an existing Fact in the same category, Company Brain flags it as a conflict. You'll see a conflict badge in the Review Queue and the Conflicts section. Resolve by approving one and superseding the other.
Facts
Facts are the atomic unit of knowledge in Company Brain. Each Fact is a single verifiable statement about your company, categorised, confidence-scored, and tied to a source.
Categories
Every Fact belongs to exactly one category, which controls how it's served to AI clients and how conflicts are detected.
| Category | What it captures |
|---|---|
| company_overview | Who you are — history, structure, founding story |
| mission | Why the company exists |
| vision | Where the company is heading |
| market | Your market, its size, and its dynamics |
| customer_profile | Your ideal customer profile |
| persona | Specific buyer or user personas |
| product | What you build |
| feature | Specific product capabilities |
| pricing | Plans, pricing, and packaging |
| positioning | How you position against alternatives |
| messaging | How you describe yourself |
| competitor | Competitors and the competitive landscape |
| workflow | How a specific piece of work gets done |
| process | Internal processes and procedures |
| policy | Company policies |
| constraint | Known limitations and constraints |
| team | People, roles, and org structure |
| ownership | Who owns what |
| metric_definition | What a given metric means |
| strategy | Goals, OKRs, and initiatives |
| risk | Known risks |
| glossary | Internal terminology and definitions |
| other | Anything that does not fit another category |
Strategy Facts
Facts in the strategy category have additional fields: strategy_type, strategy_owner, strategy_status, and review_date. You can set them when creating a strategy Fact (and they will populate automatically once OKR integrations ship), and Facts are flagged for review when a review date approaches.
Confidence Score
Every fact and proposal carries a confidence score between 0 and 1. The extraction pipeline typically assigns proposals a score in the 0.40–0.95 range based on how strongly the source supports the statement; anything weaker is kept as evidence rather than proposed. New facts default to 0.90 and proposals to 0.70. You can override any score during review.
Proposals
Proposals are the Review Queue — the holding area for candidates that need a human decision before becoming Facts. Every piece of automated intelligence lands here first.
Reviewing proposals
Open the Review Queue from the sidebar. Each card shows the proposed statement, category, confidence score, source, and supporting evidence.
Keyboard shortcuts
Bulk review
Select multiple proposals with the checkboxes and use the bulk action bar at the bottom of the page to verify or dismiss all at once. Useful after a large upload or agent run.
Conflict badges
If a proposal contradicts an existing Fact in the same category, it shows a conflict badge. Click to see which Fact it conflicts with before deciding.
Evidence & Signals
Evidence
Every approved Fact is backed by traceable evidence — the source excerpts, quotes, or data points that the extraction pipeline found. Evidence is created automatically during upload. You can also add evidence manually to any Fact from its detail panel.
Signals
Signals are observations that don't directly become Facts but are worth tracking. Brain Agents create signals when they detect market shifts, competitive moves, or internal changes. Signals with high importance scores are surfaced on the dashboard. Over time, clusters of signals can trigger new proposals.
| Signal type | What it captures |
|---|---|
| observation, insight, trend | General analytical observations drawn from your knowledge base |
| anomaly, change, contradiction | Something unexpected, a detected change, or a conflict between items |
| opportunity, risk | An opening worth pursuing, or an emerging threat |
| competitor_pricing_change, competitor_product_launch, competitor_positioning_shift, … | Competitor moves, recorded by the Competitor Intelligence agent |
| market_language_drift, market_sentiment_shift, market_emerging_frame, market_trend_signal | Market shifts, recorded by the Market Signal agent |
| tension | A strategic tension that needs resolution (Team plan) |
Data Upload
The Data Upload page is the fastest way to get knowledge into Company Brain. You can upload a file or paste text — Claude AI does the extraction automatically.
What to upload
Any text-based business document works well: company decks, strategy briefs, one-pagers, CRM exports, board updates, pricing sheets, or org charts. The extraction pipeline is optimised for dense, structured business language.
How extraction works
Every upload runs four Claude extraction passes in parallel, each targeting a different kind of knowledge:
Structured facts
Concrete, verifiable statements: names, roles, numbers, dates, product names, market definitions, pricing figures.
Strategic context
Softer, interpretive knowledge: mission, positioning language, workflow descriptions, strategic rationale, competitive framing.
Business rules
Operating logic: pricing rules, eligibility criteria, SLAs, compliance guardrails, and conditional policies.
Glossary terms
Company-defined vocabulary: product and feature names, internal jargon, and domain-specific terms.
Each pass lands its output in the Review Queue as typed proposals — facts, rules, and glossary terms — so you approve everything through the same workflow.
Supported file types
PDF, DOCX, PPTX, TXT, MD, and CSV. Maximum file size: 20 MB. You can also paste text directly.
Quality tip
Well-structured documents with clear headings and labelled sections produce significantly better extraction results than raw text dumps. A 2-page structured brief often outperforms a 20-page unstructured document.
Brain Agents
Brain Agents are autonomous background workers that keep your knowledge current without manual intervention. Enable them once — they run on a shared hourly scheduler and propose updates as your company and market evolve.
Built-in agents
| Agent | What it does | Available on |
|---|---|---|
| Digest | Summarises recent Facts, Proposals, and Signals so you can catch up on what changed in your Brain. | All plans |
| Website Intelligence | Monitors your own website for changes in positioning, messaging, pricing, and product structure. Creates Proposals when drift is detected. | Pro & Team |
| Market Signal | Monitors public channels for trends, sentiment shifts, and emerging language in your market. Creates Signals. | Pro & Team |
| Competitor Intelligence | Monitors competitor sites for feature launches, pricing changes, and positioning shifts. Creates Signals and flags risks. | Team |
Integration syncs
Each connected integration syncs automatically on a recurring schedule and creates Proposals when new or changed data is detected — HubSpot and Pipedrive every 6 hours, Slack every 2 hours, and Notion every 4 hours. Google Drive syncs when you connect it and on demand.
Agent runs
You can trigger an agent run manually from the Brain Agents page by clicking Run now. Each run is logged with duration, proposals created, and any errors. Recent runs appear in the agent's history panel.
Integrations
Integrations connect Company Brain to your existing business tools, enabling automatic knowledge extraction without manual uploads.
CRM integrations
| Integration | What it extracts | Setup |
|---|---|---|
| HubSpot | Contacts, companies, deals, pipeline stages, customer segments | OAuth or API key — connect from the Brain Agents page |
| Pipedrive | Won deals, client roster, pipeline health, deal-rot alerts | OAuth or API key — connect from the Brain Agents page |
Knowledge & collaboration
| Integration | What it extracts |
|---|---|
| Google Drive | Docs, Slides, DOCX, PPTX, PDFs — client deliverables, proposals, financial norms, presentations, templates |
| Notion | Database properties and page content — decisions, pricing norms, delivery standards mapped to Brain categories |
| Slack | Channel discussions — expertise, pricing norms, capacity signals, and team announcements |
Coming soon
More integrations are in development and will appear as connectable cards when ready — including Salesforce and Microsoft Teams, and OKR platforms (Dobee, Perdoo, Weekdone) for syncing objectives as strategy Facts.
Sync frequency
Connected integrations run incremental syncs on their own recurring schedule (see Brain Agents for the per-integration cadence). A full sync runs when you first connect an integration or change its configuration, and you can trigger a manual sync from the integration's status page.
Security
All integration tokens are stored encrypted at rest. Company Brain uses read-only OAuth scopes wherever possible. Tokens are refreshed automatically before expiry and revoked immediately on disconnect.
MCP Connect
The Model Context Protocol (MCP) is an open standard that lets AI clients fetch context from external servers. Company Brain acts as an MCP server — your AI clients query it to load your company's verified knowledge into every session.
Supported clients
Setup
Generate an API key
Go to MCP Connect and click Add connection. Select your client type and generate a key. The key is shown once — save it.
Add the MCP server to your client
Most clients (Claude.ai, Claude Desktop, ChatGPT, Cline, Microsoft 365 Copilot) add the server by URL through a connectors screen. A few (Cursor, Devin Desktop, GitHub Copilot) add it by pasting a JSON snippet into a config file. MCP Connect shows the exact steps and the correct snippet for whichever client you pick.
Test the connection
Open your AI client and ask something like "What do you know about our company?". If context loads, the connection is working.
MCP server URL
URL-based clients use this connection URL directly. Treat it like a password — anyone holding it can read your Brain.
https://mcp.company-brain.ai/mcp?key=YOUR_KEY
JSON config snippet
The config-file shape differs by client, so MCP Connect generates the exact snippet for the client you choose. For example, Cursor uses:
{
"mcpServers": {
"company-brain": {
"url": "https://mcp.company-brain.ai/mcp?key=YOUR_KEY",
"type": "streamableHttp"
}
}
}Other JSON clients differ in shape: Devin Desktop uses a serverUrl field instead of url, and GitHub Copilot uses a top-level servers key with type: "http".
Available MCP tools
Once connected, your AI client can call any of these tools during a session:
| Tool | Description |
|---|---|
| companybrain_get_company_profile | Retrieve the current company profile and metadata |
| companybrain_register_source | Register a new knowledge source for ingestion tracking |
| companybrain_list_sources | List all registered knowledge sources |
| companybrain_submit_evidence | Submit a raw evidence item for review |
| companybrain_list_evidence | List evidence items with optional filters |
| companybrain_create_entity | Create a named entity (person, product, org, etc.) |
| companybrain_list_entities | List entities with type and name filters |
| companybrain_update_entity_refs | Set or clear external cross-references (Kjerne ecosystem ID, Dybde ID) on an existing entity |
| companybrain_record_signal | Record a business signal or event observation |
| companybrain_list_signals | List signals filtered by type or date range |
| companybrain_create_proposal | Propose a new fact for human review. Write path only — does not return data for answering questions. The proposal enters a review queue and does not become a live fact until approved. |
| companybrain_list_proposals | List pending and resolved proposals |
| companybrain_review_proposal | Approve or reject a pending proposal. An approved proposal becomes a verified fact. A rejected proposal is closed with no fact created. |
| companybrain_get_fact | Retrieve a single approved fact by ID |
| companybrain_list_facts | Primary tool for retrieving verified company facts. Always filter by category when the question concerns a specific domain. Valid categories: company_overview, mission, vision, market, customer_profile, persona, product, feature, pricing, positioning, messaging, competitor, workflow, policy, constraint, process, team, ownership, metric_definition, strategy, risk, glossary, other. |
| companybrain_create_fact | Create a fact directly, bypassing the proposal queue. Requires owner or admin privilege. Write path only — does not return data for answering questions. Use companybrain_create_proposal unless explicitly granted admin rights. |
| companybrain_get_knowledge_pack | Orientation overview of the company knowledge base. Returns a category index (each fact category with its count), business rules, glossary terms, and recent changes — but not the full fact rows. Call this at the start of a session to understand what knowledge exists and which categories are relevant. Then call companybrain_list_facts filtered by category to retrieve actual facts. |
| companybrain_list_recent_changes | List recent changes to the knowledge base |
| companybrain_list_knowledge_conflicts | List detected contradictions between proposals or facts |
| companybrain_get_okr_context | Get the current OKR context — active objectives and key results, progress, alignment to facts and signals, and coverage gaps. Use for goal-setting, planning, and direction decisions. |
| companybrain_resolve_business_glossary_term | Look up the company's canonical definition of a specific term before using it in an answer. Call proactively on any company-specific terminology. Returns found=false cleanly on a miss; never an error. |
| companybrain_get_business_rules | Retrieve business rules before generating advice, commitments, or any output that could conflict with company policy. The enforcement tier (hard/soft/default) tells you how strictly each rule applies. Filter by category for focused retrieval. |
| companybrain_get_business_glossary | Browse the full glossary vocabulary for a domain. Use when you need to understand what terminology exists across a category. For a single known term, use companybrain_resolve_business_glossary_term instead. |
| companybrain_propose_business_rule | File a proposal to add a business rule for human review. Does not create a live rule — the proposal must be approved first. |
| companybrain_propose_business_glossary_term | File a proposal to add a glossary term for human review. Does not create a live term — the proposal must be approved first. |
Team Key Sharing
Team key sharing lets your colleagues connect their AI tools to Company Brain without needing a Company Brain account. One company key, shared via a link — teammates connect in under two minutes.
How it works
Generate a company key
Go to Team and click Generate Company Key. This creates a single shared key tied to your workspace.
Share the setup link
Copy the setup link (format: company-brain.ai/team-connect?key=…) and send it to colleagues via Slack, email, or any channel.
Teammates connect
When a colleague opens the link, they see your company name, pick their AI tool, and click Connect. Their setup instructions are pre-filled with the key. No login required.
Connection tracking
The Team page shows how many teammates are connected and which AI clients they're using (Cursor, Claude Desktop, ChatGPT, etc.). This updates in real time as colleagues connect.
Revoking a key
Revoking the company key immediately disconnects all teammates using it. They will need to reconnect using a newly generated key. Revoke only if you believe the key has been compromised, or to rotate access when team members leave.
Roles & Permissions
Company Brain uses six role levels. Roles are set per workspace member and control what each person can read, write, and approve. Roles are additive — each tier includes everything the tier below it can do.
| Role | Can do |
|---|---|
| Owner | Everything, including editing or deleting the workspace and transferring ownership. |
| Admin | All data operations plus integrations, agents, member invites, company MCP keys, settings, and billing. Cannot delete the workspace or transfer ownership. |
| Validator | Everything a Contributor can do, plus approve, reject, and delete proposals and view the audit log. |
| Contributor | Create sources, evidence, signals, and proposals; propose edits to Facts, rules, and glossary terms; connect integrations; manage their own MCP keys. |
| Reader | View-only access to all knowledge — Facts, proposals, signals, entities, and their own MCP keys. Cannot create or approve anything. |
| Billing Manager | Read access plus billing — view and change the plan and manage payment. No data-editing rights. |
Inviting teammates
Go to Invite Others from the sidebar. Enter the email address and select a role. The invited person receives an email with a unique link. Links expire after 7 days.
Company key vs. workspace membership
Teammates who connect via the company key (Team page) are not workspace members — they only get read access to your knowledge via MCP. They cannot log in, review proposals, or manage anything. For full workspace access, invite them via Invite Others.
Plans & Pricing
Company Brain is available on three plans. All plans include the core knowledge engine — facts, proposals, evidence, signals, glossary, business rules, and MCP access. Higher plans unlock more members, capacity, and advanced agents.
Free
$0 /mo
- 1 workspace member
- Unlimited sources
- 50 facts
- Unlimited proposals
- Digest Agent
- MCP Connect & team key sharing
Pro
PopularPro see billing
- 5 workspace members
- Unlimited sources
- 2,000 facts
- 2,000 proposals / month
- Website & Market intelligence agents
- Integrations: CRM, Slack, Notion, Drive
Team
Team see billing
- 25 workspace members
- Unlimited sources
- 5,000 facts
- 10,000 proposals / month
- All agents incl. Competitor Intelligence
- Priority support
Upgrade your plan
To upgrade, go to Billing from the user menu at the bottom of the sidebar. You can view your current plan, usage, and upgrade at any time.
| Feature | Free | Pro | Team |
|---|---|---|---|
| Workspace members | 1 | 5 | 25 |
| Knowledge sources | Unlimited | Unlimited | Unlimited |
| Facts | 50 | 2,000 | 5,000 |
| Proposals / month | Unlimited | 2,000 | 10,000 |
| Digest Agent | ✓ | ✓ | ✓ |
| Website Intelligence | — | ✓ | ✓ |
| Market Signal Agent | — | ✓ | ✓ |
| Competitor Intelligence | — | — | ✓ |
| Integrations (CRM, Slack, Notion, Drive) | — | ✓ | ✓ |
| Team key sharing | ✓ | ✓ | ✓ |
| Priority support | — | — | ✓ |
Business Glossary
The Business Glossary is your company's canonical dictionary — a curated set of terms, definitions, synonyms, and scope boundaries that give AI tools a shared vocabulary for understanding your business.
Without a shared glossary, AI tools make assumptions. "ICP" might be interpreted as a generic marketing concept rather than your specific customer definition. "Activation" might follow a textbook formula rather than the metric your team actually measures. The glossary closes that gap by injecting your definitions into every AI session via the MCP knowledge pack.
What a glossary term contains
| Field | Purpose |
|---|---|
| Term | The canonical name for the concept — what your team actually calls it |
| Definition | A precise, agreed-upon explanation of what the term means in your context |
| Synonyms | Alternative names or abbreviations used for the same thing, so AI recognises them too |
| Category | Groups the term by domain: product, market, strategy, process, metric, etc. |
| Boundary note | Clarifies what the term does not cover — reduces misapplication |
| Confidence score | How certain Company Brain is about the definition (0.40–0.95) |
How terms are added
Terms can be added manually via the Glossary page, proposed by Brain Agents during fact extraction, or surfaced when the extraction pipeline encounters a term it cannot resolve. Terms marked Needs definition have been recognised as important but not yet defined — they appear as a work queue on the Glossary page.
Editing terms
Changes to glossary terms go through the proposal workflow — edits create a new versioned record pending review, so the canonical definition is never overwritten without an approval step. The previous version is retained and traceable.
Glossary and AI context
When an AI tool calls the companybrain_get_business_glossary MCP tool, it receives your full active glossary. The companybrain_resolve_business_glossary_term tool lets AI resolve a specific term by name or synonym, including fuzzy matching.
Business Rules
Business Rules are explicit, machine-readable statements that capture how your company operates — what is required, what is forbidden, what is permitted, and what depends on conditions. They give AI tools guardrails so they can reason within your actual operating constraints, not generic assumptions.
Rule anatomy
| Field | Purpose |
|---|---|
| Statement | The rule itself, written as a clear declarative sentence |
| Direction | Requires · Forbids · Permits · Conditional — how the rule applies |
| Enforcement tier | Hard (absolute) · Soft (strong default) · Default (normal practice) |
| Condition | Optional: when the rule applies — "when dealing with enterprise customers", "in regulated markets" |
| Rationale | Why the rule exists — helps AI understand intent, not just the constraint |
| Category | Groups the rule by domain: pricing, process, legal, product, hiring, etc. |
Enforcement tiers
Absolute. No exceptions. Used for legal obligations, regulatory constraints, or non-negotiable company policy.
Strong preference. AI should follow by default but can deviate when context clearly justifies it.
Normal practice. Describes how things are typically done. Lowest friction to override.
How rules create value
A well-maintained rule set means your AI tools know not to quote custom pricing without approval, not to promise delivery timelines that violate SLA policy, and to always recommend legal review for contracts above a certain threshold — without anyone having to prompt them. Rules shift AI behaviour from generic to company-specific at inference time.
Adding and editing rules
Rules are added via the Business Rules page. Like glossary terms, edits go through the proposal workflow — every change creates a new versioned record pending review, preserving the audit trail.
Rules and AI context
The companybrain_get_business_rules MCP tool returns your full active rule set to any connected AI client. Rules are included in the knowledge pack alongside facts and glossary terms, so they inform AI reasoning automatically — no manual prompting required.
Still have questions?